This is one of several design articles about the new interactive narrative platform Versu, which Richard Evans and I have been building with a team at Linden Lab.
The first several stories implemented for Versu are written for a generally Regency-period sensibility, influenced heavily by Jane Austen, and less-heavily by several of her contemporary authors, as well as more recent writers such as Georgette Heyer and Patrick O’Brian. We did not pull quotes from the more recent authors whose work is still in copyright, but we were aware of some of the tropes that those authors have brought into reader consciousness.
That wasn’t at all an inevitable choice. Versu as a platform is up to representing lots of possible periods, genres, and sub-cultures, and the next content we release will introduce material from another genre, so I’d like to talk a little about why we started where we did.
We chose Austen’s world as a starting point for Versu because — though the system is up to all kinds of possible content and genres — Austen’s work takes place in a world of precisely defined manners that is nonetheless fairly familiar to contemporary readers.
The focus on manners, and the relatively large body of prose all referring to the same social milieu, meant that Austen gave us a lot of different examples of characters reacting to one another, speaking with subtext, and following (or failing to follow) implicit social rules. This made for terrific sample material on which to test and robustify Versu’s system. We reasoned that if we could make something strong enough to account for many Austen situations, we would be able to use that same system to describe the wide range of other socially focused stories we wanted eventually to be able to tell.
Other highly mannered societies would also have been great fun to model — Heian Japan came to mind. But it seemed like a dangerous idea to start with something where the average reader would have so very much to learn about the society in question just in order to anticipate the results of her actions. We wanted a reader’s first encounter with Versu to be one in which she experienced a sense of agency because she did know how to influence other characters. (For the same reason, the second Versu genre is a modern office setting — again, to establish accessibility.)
Down the line, we can imagine Versu being used educationally to do precisely this — introduce a particular historical period with its unique concerns, allow students to create models that represent the results of their research, or teach players about social expectations in specialized contexts. We can also imagine stories about alien worlds with vastly different manners that the reader needs to learn to navigate: socially focused speculative fiction.
From Austen, we drew a lot of particulars of what to model in a shared genre-specific library. We read through Austen’s novels, unfinished works, and letters, and extracted information to feed into Versu’s model.
In Austen, what do characters evaluate about one another? About themselves? Propriety, attractiveness, intelligence, and social status came up a lot.
What relationships are important? Various stages of romantic connection, from Frank Churchill and Emma’s light flirtation to Jane Fairfax and Frank Churchill’s serious (but secret) engagement came into play; family relationships were important, and so was the question of who was destined to inherit from whom.
Marking up Austen this way also turned our attention to some things that the Versu model didn’t originally include, or didn’t handle in a sufficiently flexible way. Austen characters spend quite a lot of time speaking explicitly about their values and beliefs, from Lady Catherine pontificating about the importance of musical accomplishment to Elizabeth talking about whether she thinks a man’s poverty would be enough to dissuade her from marrying him. From the outset, we had modeled the idea that different characters would care about different issues, so one character might be status-conscious while another was more susceptible to a pretty face. But we hadn’t modeled the ability for characters to explicitly discuss these values, argue about them, and change their minds. Drawing from Austen, we added dialogue elements that handled these types of conversations; and characters who, if persuaded to change their minds about a particular value, would then act slightly differently as a result. Characters in Versu who care strongly about propriety will have a different set of available reactions than ones who don’t care about it at all, for instance.
Several anecdotes from Austen’s letters became fodder for dialogue, such as her description of a storm that damaged several of the trees in the neighborhood. The letters also supplied a number of generalized moral and ethical statements that we pulled as dialogue when characters are talking about their beliefs.
From Austen and Austen-related research materials we also derived a lot of information about common contemporary entertainments and social scenarios, which set up the background for scenes in which characters are getting to know one another or resolving social states. There are a number of songs that characters can perform in the Regency Versu scenes, for instance, whose lyrics are drawn straight from Jane Austen’s songbook. Fordyce’s Sermons, mentioned in Pride and Prejudice, becomes a text from which characters may read aloud genuine snippets.
Sometimes these researched treatments surprised us a bit: for instance, the order of presenting dishes at a Regency dinner was quite a bit different from the course-by-course handling we’re used to. Far from starting with a soup or salad course and working through fish and meats, Regency tables often featured a wide selection of dishes simultaneously — so that’s what we described in “A Family Supper.”
How NOT Austen?
One of our early playtesters reacted with a pretty harsh “hey, I love Austen, and this IS NOT AUSTEN” response. And that’s a completely fair remark. There are a lot of things about Austen that we could not emulate or chose not to include.
Prose Cadence and Viewpoint
Perhaps the most obvious is the texture of Austenian prose. Austen writes a lot of very long paragraphs, and frequently summarizes the events of long periods of time, or overall impressions that characters take away from multiple encounters. Because we wanted to focus on moment-to-moment action, we deliberately moved away from Austen’s prose cadence. We were more invested in the idea that the player would be allowed to make large or small social moves: it was that Austenian sense of cumulative small details as well as grand gestures that we focused on most.
Similarly, Austen has a very distinctive style where she subverts the supposedly dispassionate third-person and uses it to reveal a character’s interiority. The very first sense of Pride and Prejudice (“It is a truth universally acknowledged…”) is not a truth universally acknowledged — it is from the perspective of a particular character: Mrs Bennett.
(Richard notes: This technique is not just a cheap stylistic trick, but is arguably [according to Bharat Tandon at least] at the heart of her vision: there is no dispassionate third-person external view from nowhere: we are always already seeing the situation from a particular, value-laden perspective, always already engaged, committed, etc.)
We don’t do this at all in Versu. Once we committed to the play-format, we couldn’t. Committing to the play format was the right decision for us for a ton of reasons, but it does mean we can’t do free indirect style.
Despite the research that came into play, history fans and Austen buffs will find many many ways in which we ignore or contravene period conventions. Versu’s Regency genre file doesn’t, for instance, rigorously enforce the different titles characters use for one another. Properly speaking, characters in Austen address one another in ways that are very much dependent on their relationship states, with the use of first names typically being acceptable only between intimates. In practice, what we found was that trying to use this system uniformly made gameplay confusing: now the reader was expected to track interactively not only all the characters in the scene, but also all the things they might call one another. So we dialed this back. It is possible to specify direct address methods between two characters, so that a lady might address her husband as “my dear,” but Versu doesn’t aggressively manage this.
Another point is that Austenian plots tend to focus heavily on character problems rather than on striking events. What will happen between Elizabeth and Darcy? Will Anne Elliot ever reconcile with Captain Wentworth? Can Fanny Price ever obtain happiness, when her horizons are so limited?
These are terrific stories, but a literal adaptation of them makes for directionless gameplay, especially in the opening moves. We found that just setting the reader down in a ballroom with some emotional malaise didn’t make for strong engagement. Instead, we needed something where the protagonist’s problem was more clearly defined, more immediate, more demanding.
That’s why we pulled in mystery and gothic plot elements for two of our longer stories, even though Austen herself engaged with the gothic primarily in order to make fun of and subvert it, in Northanger Abbey. We probably could have explored a more Austenian genre-subversion, but we felt that at the beginning of the project, genre expectations were positively helpful to us, because they give the player hints about what to expect to be able to accomplish in the story.
(And to be fair, even Austen sometimes made use of a strong precipitating incident — the unfinished Sanditon begins with an overturned carriage and a wounded protagonist.)
Another significant playability point is that Austen’s characters practice quite a bit of deception, from Frank Churchill’s fake flirtation with Emma to Edward Farrars’ silence about his engagement. This makes for fun reading, and to a limited extent Versu characters can also practice deception. But we had to tread a fine line here. Versu doesn’t (currently) implement very deeply the question of what characters believe about the knowledge states of other characters — this quickly becomes a very very complicated and recursive business — and that ruled out certain treatments of deception. We were also concerned that if characters did too much lying and other deceptive behavior, perfectly correct simulator output would look like a mistake to the player, because characters would be responding based on information the player didn’t have.
This question of character deception is one I’d love to come back to and address more systematically, but we felt that both from a modeling and a gameplay perspective, we first needed to nail the honest social interaction, reach a point where the player had a strong sense of social agency, and earn some reader trust. Only after that will we be ready to model complex secrets and lies.
Austen’s world is rigorously heteronormative, and all her characters are presented as straight. We wanted to avoid building those assumptions into the engine. At the same time, we wanted to avoid startling readers who were trying to keep to canon by having an AI-driven NPC Mr Darcy unexpectedly hook up with an AI-driven NPC Mr Bingley, contrary to their canonical behavior.
So the compromise we settled on was this:
- characters can be defined to be interested in zero, one, or multiple genders (and the system is able to acknowledge genders besides male and female, though there are no Austen characters who embody this)
- a character driven by a player can ignore their character’s pre-defined sexual preference and make advances to others; so a player controlling Elizabeth may make advances to another woman
- a character driven by the AI will by default adhere to their specified preference, so an AI-driven Elizabeth will not make advances to another woman by default
- if a player character makes advances to an AI character, that will automatically make that AI character open to the player character’s gender
- any characters who can reach a state of strong romantic involvement may become engaged, though engagement and marriage are not acknowledged for non-straight couples in Regency England; the only way this can happen in the simulator is for a player character to take action in that direction
There are still a lot of situations this doesn’t cover: it’s obviously possible that an author will want to write a story about the protagonist’s fruitless crush on a person who isn’t interested in her gender, for instance. Some ways to address this are in the works.
A related point is that there’s a lot of gender-role policing inherent in the manners of the Regency. Men are expected to behave one way, women another, and characters tend both to talk about these differences in essentialist ways, but also to give strong feedback to those who stray beyond their “appropriate” gender presentation.
Here again we took a route of trying to constrain the AI characters but not necessarily the player. Non-player women can’t challenge one another to duels; player women can, if they choose.
Another 19th-century vs. 21st-century challenge came from the treatment of mental illness in “House on the Cliff.” The mad person is a staple of the gothic novels from which we were pulling tropes. On the other hand, 19th century treatments of mental illness don’t align very well with our 21st century knowledge of those conditions, and a growing desire not to stigmatize them or spread misinformation.
The compromise we made here — and I don’t know whether this was the right approach or not, honestly — was to try to make the story’s “madness” as obviously trope-madness as possible: a representation of a story feature from a particular period and genre, not a representation of the way mental illness works in the real world. “House on the Cliff”‘s “mad” people contract madness almost as a Lovecraftian disease.
In all of these cases, I’m very conscious that we’re walking an uncomfortable line, and it’s because there are so many different ways to engage with the mores of the past:
- represent accurately and uncritically (and even this raises some issues: Austen’s omission of gay characters obviously doesn’t mean that no one in the Regency was gay, so “Austen’s Regency” and “the actual Regency” are different creatures — which do we want to present “accurately”?)
- represent accurately but with criticism (e.g. by showing negative outcomes of historical expectations)
- represent inaccurately, treating the period as a fantasy dressing for some subset of modern mores (since “modern mores” doesn’t even refer to a single set of beliefs)
…and ideally I would like Versu to be a platform that allows people the full creative spectrum — to fantasize, to create escapist literature in which their Mary Sue self seduces Mr Darcy, to build crossovers and slashfic to their heart’s content; to create fun experiences for their friends, grounded in in-jokes and private connections; to write serious critiques of the past or the present; to put the player in a hard situation and use that to teach a particular, personal reality; to share what their own lives are like.
We’re a long, long way from handling all of those possible uses of the system — but that’s part of the reason that user content is so vitally important. And is why the next thing we’re working on is opening up the user tools.
12 thoughts on “Versu and Jane Austen”
I wonder if we can give these a catchy monicker – e.g. the Five Laws of Gender-Fluid Robotics. Something like that:
1. characters can be defined to be interested in zero, one, or multiple genders (including genders besides male and female)
2. a character driven by a player can ignore their character’s pre-defined sexual preference
3. a character driven by the AI will by default adhere to their specified preference
4. if a player character makes advances to an AI character, that AI character becomes open to the player character’s gender
5. any characters who can reach a state of strong romantic involvement may…….. become engaged
I like these (and, certainly, violating gender politics was the first agency I reached for in your Austen world)
I imagine they must have been hell to program, though…
That’s a pretty good set of base assumptions in a sandboxy thing where very open player choice is a central motive, yeah.
A question that’s come up whenever I’ve talked to people about Versu has been ‘can it do anything other than comedy of manners?’ It is surprisingly difficult to get across to people that, outside the funky world of computer games, the great majority of stories that have ever been told are mostly about social interactions.
I don’t understand why Verso cannot produce statements like. “It is a truth universally acknowledged…” or engage with the sort of social critique that Austen does. This seems like exactly the kind of work Verso could do and do quite well.
Consider the first lines of Pride and Prejudice:
“It is a truth universally acknowledged, that a single man in possession of a good fortune must be in want of a wife.
However little known the feelings or views of such a man may be on his first entering a neighbourhood, this truth is so well fixed in the minds of the surrounding families, that he is considered as the rightful property of some one or other of their daughters.”
These statements aren’t free indirect speech on the part of Mrs. Bennet, they are narration provided by the narrator. The narrator isn’t describing an interior, psychological state of a single person from the inside, but is making an external observation about a mindset common to a particular (not universal) social environment (a neighborhood), that is organized by an fixed idea understood to be the truth (single, white men with money are property), a truth upheld by persons within the environment (surrounding families), who will collectively and forcefully determine the actions (marriage, children) of any well-to-do single man who enters its domain. In other words, this is a socially conformist environment, ruled by manner and convention, in which an entering, eligible man only has one option, which can hardly be called a choice, for once he enters the neighborhood, he becomes property. If it helps, think Stepford.
This seems like a world model (with many possible character models) that is completely amenable to Verso, without loss of Austen’s social critique. In these opening sentences, Austen reverses a major, commonsense or ideological presumption of the British landed gentry in the early nineteenth century — that single, white men with wealth were sovereign, free, reasonable, intentional, conscious, ruling subjects. Instead, Austen portrays them as environmentally determined objects. They are portrayed as possessive individuals who are missing something (a wife) and this lack turns them into property (possess-able individuals), the use of which is absolutely determined by a historically specific, social milieu who rule through drawing room manners deployed primarily by women. Austen turns the world-view of the British Empire inside out in two sentences. Verso can too.
To be clear, I was not saying that Versu can’t output Austen’s opening sentence — it can certainly print hand-authored chunks of text — or that it can’t do social critique by means of its world and social model. I believe it certainly can, which is part of the purpose of building such an engine in the first place.
Indeed, there’s an obvious implicit critique embedded in the system as it currently stands: characters marked as being of the “servant” class lose access to many social practices and aren’t allowed to say certain things or visibly react to the antics of their masters.
In the available content at the moment those servant roles are simply unavailable to the player on the grounds that we didn’t want that frustrating experience — of being a bit participant, endlessly clearing dishes, serving tea, and keeping silent, being limited to answering questions and gossiping only with other servants — to be someone’s first contact with Versu. But if one actually plays one of the servant roles, the experience becomes a dramatic representation of how much those characters are silenced and regimented. It would be easy to imagine fleshing out one or more of the servant characters with internalized coping mechanisms: internal thoughts about their horrid masters, perhaps, or subversive dialogue available only in the presence of other servants, etc.
Anyway, what we were getting at in this article was rather that the engine isn’t procedurally generating long prose paragraphs of indirect narration or summary of social attitudes. It manifests character choices through short-term tasks, rather than by trying to convey generalities of how they might interact over long periods. And even if the engine were able to come up with some kind of summary data, it’s not sophisticated enough in its procedural text generation to describe the results.
Characters themselves can have things to say about their social attitudes, though, and can discuss these with one another.
Yes, I see that. Sorry, I could have made this point much more succinctly. Short, introductory frames and clever word choice are a staple of IF. Austen’s two-sentence introductory frame and her word choice engage in a style of fiction writing that is not contrary to IF.
Specifically, word choice in Austen’s two-sentence frame uncovers the economic foundation of society, through the use of words that have both an economic and social valence. The critique in her novels depends, in part, on the wording of an initial frame and the continued repetition of economic terms to describe social mores. The Bennets may recognize and identify with the social meaning (marriage as continuity of kinship) of the opening lines, but not the economic meaning (marriage as the continuity of property/capital). Austen’s reader can understand both. In Versu, an interactor may have to chose, repetitively, either to act as a Bennet (to engage in a manners play and not respond to its underlying, determining economy) or to act as an interactor (who is conscious of how manners cover the working of economy).
What produces dramatic irony in a play or a novel, when used in IF, makes the interactor, like the reader, aware of the difference between what the characters say and what their words might mean, economically, ethically, ideologically, etc. Then, it asks the interactor, unlike the reader, to immediately make decisions from this place of insight.
This is what I meant to convey, and failed to, earlier.
Sorry, spellcheck error – Versu.
I remain a little disappointed that this is only available through the appstore. Alright; a lot disappointed.
I fail to understand why this program requires an always-on internet connection. Isn’t this a single-player experience? I am already concerned about the cost aspect of this connection, but should I also worry about my privacy (background data collection and such)?
should I also worry about my privacy (background data collection and such)
Not to the best of my knowledge. Versu is using the server to provide the AI muscle.
But we do hear the concerns that constant connection is an issue for some people and rules out some desired uses.